About Magic Layers

Magic Layers takes a single finished image — a poster, a thumbnail, a slide screenshot, a product shot — and returns its major visual elements as separate images. You get a generated base image plus one transparent PNG per element, each with a name, a short description, and the position it occupied in the original. It is built for the common case where a visual exists but its project file does not: the source was never shared, was lost, or was produced by someone else. The separation is inferred from the pixels alone.

1. You upload one PNG or JPEG, at least 512×512 pixels; larger images are downscaled automatically to at most 10 MB. It goes to private storage, and only a short-lived signed URL is passed on. 2. You choose the target count — from 2 to 15 images per job, counted as one base image plus the rest as layers — and describe what to separate, for example “all people in the image”, or leave it blank for a general split. 3. A third-party AI model (ByteDance Volcano Engine Ark / Seedream) analyses the image and produces the separated elements. 4. You review the layers, see where each one sits in the original, and download the ones you need. The site runs on Cloudflare Workers, with D1 for account and billing records and R2 for temporary image storage.

Being straightforward about the limits saves everyone time: • The layers are transparent raster PNGs, not a recovered Photoshop, Figma, or Illustrator document. Live text, fonts, vector paths, and effects that only existed in the authoring file cannot be brought back from a flat image. • The target count is a request, not a guarantee. A simple image may contain fewer meaningful elements than expected — in which case you get fewer images and pay less. • Boundaries, names, and quality vary by image. Busy or low-contrast compositions separate less cleanly than graphic, high-contrast ones. • Some images are refused by the processing provider — for example copyright-protected or adult content. A refused image fails before billing — no credits are charged. • Separation is a technical operation, not a licence. It does not grant you any rights in the source artwork.

Uploaded source images are private to your account and are never published or shown to other users. Source-image objects are configured for seven-day retention. Generated layers are returned directly by the processing provider for immediate preview and download and are not copied into Magic Layers storage. Full detail is in the Privacy Policy and the Terms of Service.

Usage is billed in Credits. Each output image costs 2 Credits, including the base image. You choose the target count (2–15 images per job) and need a balance of at least 2 Credits per selected image to start. After each job you are billed for every image actually returned — the model may return fewer, or slightly more than your target, and over-delivery can take your balance below zero. A negative balance must be topped up before the next job. Failed jobs are never charged. Credits come from a monthly plan or a one-time Credit Pack. Current prices are on the Pricing page.

Magic Layers is an independently built and operated web product. It is not affiliated with, endorsed by, or a partner of Adobe, Figma, Canva, or any other tool named on this site; those names appear only to describe where the exported layers can be used.

Questions, bug reports, billing problems, and takedown requests all go to contact@magiclayers.net. It is a monitored address, not a no-reply.